Mining frequent patterns and their evolution in continuous-time temporal networks

Ali Jazayeri

doi:10.17918/00000919

Networks are used as highly expressive tools in different disciplines. In recent years, the analysis and mining of temporal networks have attracted substantial attention. Frequent pattern mining is considered an essential task in the network science literature. In addition to the numerous applications, the investigation of frequent pattern mining in networks directly impacts other analytical approaches, such as clustering, quasi-clique and clique mining, and link prediction. Many of the algorithms proposed for frequent pattern mining in temporal networks represent these networks as sequences of static networks. Then, the inter- or intra-network patterns are mined. This type of representation imposes a computation-expressiveness trade-off to the mining problem. In this dissertation, we propose a novel representation that can preserve the temporal aspects of the networks losslessly. Then, we introduce the concept of constrained interval graphs (CIGs). We develop a series of algorithms for mining the complete set of frequent temporal patterns in a temporal network data set based on this concept. We also consider four different definitions of isomorphism to allow noise tolerance in temporal data collection. Implementing the algorithms for three real-world data sets proves the practicality of the proposed algorithms and their capability to discover unknown patterns in various settings. We also propose an approach for mining frequent evolving patterns. We develop a series of algorithms for the detection of different evolution events in continuous-time temporal networks. The events detected and their relationships are used to create evolution networks representing the evolution of the corresponding temporal networks over time. Then, the frequent evolving patterns are mined from the data set of evolution networks. We apply the proposed algorithms to three real-world data sets. The findings show different frequent evolving patterns are common in various settings. Sepsis is one of the most challenging health conditions worldwide, with relatively high incidence and mortality rates. It is shown that preventing sepsis is the key to avoid potentially irreversible organ dysfunction. However, data-driven early identification of sepsis is challenging as sepsis shares signs and symptoms with other health conditions. To further study the usefulness of the proposed approaches in this dissertation, we apply the developed algorithms to a data set of sepsis patients. We show that using frequent patterns identified by the proposed algorithms as features for classifying sepsis and non-sepsis patients can improve prediction accuracy and performance. Most of the temporal modeling approaches adopted in the sepsis literature are based on deep learning methods. Although these methods produce high accuracy, they generally have limited model explainability and finding interpretability. Using the adopted methods in this study, we could identify the most important features contributing to the patients' sepsis incidence and enhance the clinical interpretability. The findings show that the frequent temporal and evolving patterns identified by the proposed algorithms can be directly used to analyze common interactions in different settings or as features for other machine learning algorithms.

Mining frequent patterns and their evolution in continuous-time temporal networks

Files and links (1)

Abstract

Metrics

Details

Mining frequent patterns and their evolution in continuous-time temporal networks

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media